Stanford Med · Unit 5B Medicine·Day shift·Mon · 9:42 AM

AI agents

The Claude agents Throughline runs · live activity from the audit spine

Every agent is a typed function over a pinned Anthropic model with a Zod-validated output schema. Read-only agents run automatically; HITL-gated agents draft a proposal that no integration touches until a licensed reviewer approves — that click writes the Submission audit row. Fire a scenario beat below and watch the agents react in real time.

Scenario beats · Michael Johnson

Fire an event and watch the agents react

Live agent activity

Agents reacting to the unfolding case

0 events
No agent activity yet. Fire a scenario beat above and watch the agents react — read-only agents run; HITL-gated agents draft a proposal for the queue below.
Activity reflects deterministic orchestration — agents reasoning over the engine, never inventing rules. ~figures are each agent's typical per-call cost. A real runAgent call (real cost + AgentInference row) and the Submission write happen at the HITL gate when a reviewer approves.
Execution layer · HITL queue

Pending actions · awaiting clinician approval

0 open

Drafted by playbooks / AI · NOT sent until you approve. Approve fires the integration mock + writes a Submission audit row.

No pending actions. Click Run playbook on a barrier to draft one.

Live agents

Implemented and audit-logged via runAgent()

Planned agents

Declared in the AgentName union; prompts + tool schemas pending. Listed for transparency, not as shipped capability.

Per CLAUDE.md: every agent call goes through runAgent in src/lib/ai/audit.ts — no exceptions. The Cockpit Copilot uses the Anthropic tool-use loop directly, but still writes one AgentInference row per chat turn.

Demo data · no PHI · mocked Epic + payer endpoints